Move model provisioning policies out of the simulator

.flake8

@@ -3,3 +3,4 @@ max-line-length = 120
 # Ignore E402: module-level import not at top of file
 # Ignore W503: line break before binary operator (incompatible with W504)
 ignore = E402,W503
+exclude = .venv

.gitignore

@@ -11,6 +11,9 @@
 *.sln.docstates
 *.env
 
+# Environment files
+.venv/
+
 # User-specific files (MonoDevelop/Xamarin Studio)
 *.userprefs
 

simulator/__init__.py

@@ -0,0 +1,15 @@
+"""
+Simulator package — provisioning sweeps, multi-request analysis, and plotting
+on top of the model_provisioner allocation policies.
+
+The allocation policy implementations live in ``streamwise/model_provisioner/``.
+"""
+import os
+import sys
+
+# Make model_provisioner importable for simulator modules.
+_STREAMWISE_DIR = os.path.normpath(
+    os.path.join(os.path.dirname(os.path.abspath(__file__)), "..", "streamwise")
+)
+if _STREAMWISE_DIR not in sys.path:
+    sys.path.insert(0, _STREAMWISE_DIR)

simulator/actions.py

@@ -27,7 +27,7 @@
 from sim_types import Objective
 from sim_types import Policy
 
-from policies import STREAMWISE_POLICY
+from model_provisioner.policies import STREAMWISE_POLICY
 
 from models import get_model_allocation
 

simulator/auto_model_allocator.py

@@ -19,7 +19,7 @@
 from sim_types import GPUType
 from sim_types import Result
 
-from policies import STREAMWISE_POLICY
+from model_provisioner.policies import STREAMWISE_POLICY
 
 from model_allocator import ModelAllocator
 
@@ -47,39 +47,39 @@ def __init__(
     def _build_allocator(self) -> ModelAllocator:
         """Create concrete allocator based on configured solver."""
         if self.policy.solver == Solver.GREEDY:
-            from greedy import GreedyAllocator
+            from model_provisioner.greedy import GreedyAllocator
             return GreedyAllocator(
                 workflow=self.workflow,
                 latency_data=self.latency_data,
                 power_data=self.power_data,
                 policy=self.policy,
             )
         if self.policy.solver == Solver.NAIVE:
-            from naive_baseline import NaiveAllocator
+            from model_provisioner.naive_baseline import NaiveAllocator
             return NaiveAllocator(
                 workflow=self.workflow,
                 latency_data=self.latency_data,
                 power_data=self.power_data,
                 policy=self.policy,
             )
         if self.policy.solver in {Solver.GUROBI, Solver.HIGHS}:
-            from milp import MILPAllocator
+            from model_provisioner.milp import MILPAllocator
             return MILPAllocator(
                 workflow=self.workflow,
                 latency_data=self.latency_data,
                 power_data=self.power_data,
                 policy=self.policy,
             )
         if self.policy.solver == Solver.HEXGEN:
-            from hexgen import HexGenAllocator
+            from model_provisioner.hexgen import HexGenAllocator
             return HexGenAllocator(
                 workflow=self.workflow,
                 latency_data=self.latency_data,
                 power_data=self.power_data,
                 policy=self.policy,
             )
         if self.policy.solver == Solver.HELIX:
-            from helix import HelixAllocator
+            from model_provisioner.helix import HelixAllocator
             return HelixAllocator(
                 workflow=self.workflow,
                 latency_data=self.latency_data,

simulator/data_loading.py

@@ -28,15 +28,17 @@
 from constants import POWER_GPU_IDLE
 from constants import POWER_GPU_TDP
 
+_DEFAULT_DATA_DIR = Path(__file__).resolve().parent / "data"
+
 
 def load_latency_data(
-    data_dir: str = "data/",
+    data_dir: str | Path = _DEFAULT_DATA_DIR,
 ) -> LatencyData:
     """
     Load latency and throughput mapping data from CSV files.
 
     Args:
-        data_dir (str): The directory where the CSV files are stored.
+        data_dir: The directory where the CSV files are stored.
     Returns:
         LatencyData: An object containing all loaded latency data.
     """
@@ -107,13 +109,13 @@ def load_latency_data(
 
 
 def load_power_data(
-    data_dir: str = "data/"
+    data_dir: str | Path = _DEFAULT_DATA_DIR
 ) -> PowerData:
     """
     Load power consumption data from CSV files.
 
     Args:
-        data_dir (str): The directory where the CSV files are stored.
+        data_dir: The directory where the CSV files are stored.
     Returns:
         PowerData: An object containing all loaded power consumption data.
     """
@@ -216,7 +218,7 @@ def load_power_data(
 
 
 def load_adaptive_quality_data(
-    data_dir: str,
+    data_dir: str | Path,
     level: QualityLevel,
 ) -> LatencyData:
     """Load latency data for adaptive quality."""

simulator/model_allocator.py

@@ -27,7 +27,7 @@
 from models import UpscalerModelAllocation
 from models import OthersModelAllocation
 
-from policies import NAIVE_POLICY
+from model_provisioner.policies import NAIVE_POLICY
 
 
 class ModelAllocator(ABC):

simulator/multirequests.py

@@ -18,7 +18,7 @@
 
 from workflows import PODCAST_WORKFLOW
 
-from policies import STREAMWISE_POLICY
+from model_provisioner.policies import STREAMWISE_POLICY
 
 from auto_model_allocator import AutoModelAllocator
 

simulator/provisioning.py

@@ -33,7 +33,7 @@
 
 from auto_model_allocator import AutoModelAllocator
 
-from policies import STREAMWISE_POLICY
+from model_provisioner.policies import STREAMWISE_POLICY
 
 from constants import SECONDS_IN_HOUR